|Publication number||US6385580 B1|
|Application number||US 09/381,630|
|Publication date||May 7, 2002|
|Filing date||Mar 20, 1998|
|Priority date||Mar 25, 1997|
|Also published as||DE69817550D1, DE69817550T2, EP0970467A2, EP0970467B1, WO1998043236A2, WO1998043236A3|
|Publication number||09381630, 381630, PCT/1998/507, PCT/SE/1998/000507, PCT/SE/1998/00507, PCT/SE/98/000507, PCT/SE/98/00507, PCT/SE1998/000507, PCT/SE1998/00507, PCT/SE1998000507, PCT/SE199800507, PCT/SE98/000507, PCT/SE98/00507, PCT/SE98000507, PCT/SE9800507, US 6385580 B1, US 6385580B1, US-B1-6385580, US6385580 B1, US6385580B1|
|Inventors||Bertil Lyberg, Mats Wiren|
|Original Assignee||Telia Ab|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (11), Referenced by (11), Classifications (22), Legal Events (10)|
|External Links: USPTO, USPTO Assignment, Espacenet|
1. Field of the Invention
The present invention relates to reproduction of speech and to the speech belonging facial movements of a speaking person. The invention is intended to be utilized in connections where a person s facial movements shall be reproduced simultaneously with produced sound.
2. Discussion of the Background
At speech synthesis there is a need to synchronize the speech with the facial movements of a speaking person. In patent application No. 9504367-5 is described how movement patterns in a face are recorded and stored together with a polyphone collection (sounds) at concatenation synthesis based on half-syllables. The recording of the movement patterns of the half-syllables of the subject (person) after that influences points in a polygon model of the face. Another texture, i.e. another face can be applied on top of the polygon model and at that get lip and facial movements from the polyphone synthesis.
With the described model is required that voices of men, women and children are recorded separately. Such procedures are expensive and circumstantial.
The present invention, and with reference to FIG. 3, relates to a method at speech synthesis for reproduction of facial movements of a person who has been allocated a speech via speech synthesis. Said speech is put together of polyphones which are fetched from a database (step S1). A databank is further established containing polyphones with to the polyphones belonging movement patterns in the face of a first person (step S2). Polyphones from a second person are further registered and stored in a database (step S3). The sound segments in corresponding polyphones in the databank and the database are compared (step S4), and the facial movements in the databank are modified in accordance with the deviation (step S5). The modified movement patterns are stored in the database and are related to the polyphone in question (step S6). The registered polyphones are after that utilized for putting together words and sentences at the same time as corresponding movement patterns build up a face model from the movement patterns in the database (step S7).
Speech from a subject (person) is recorded at the same time as the movement pattern of the subject is registered. The recorded speech preferably consists of nonsense words from which polyphones, half-syllables, are sorted out. The registered polyphones are stored in a polyphone bank. To each polyphone is further stored in a movement bank the facial movements of the subject. For a second person is in corresponding way polyphones registered in a polyphone base. The second person's facial movements, however, are not registered. A comparison between the sound segments in corresponding polyphones is after that made between the polyphone base and the polyphone bank. The registered differences are after that utilized to modify current movement pattern in the movement bank, at which a model with a movement pattern corresponding to the second speaker's pronunciation of the polyphones is obtained. The modified movement pattern is stored in a movement base. At putting together polyphones from the polyphone base, the movement base is after that utilized for creating of a face model, the movements of which correspond to the speaking person's way of speaking. The created model consists of a polygon model based on the movement pattern from the movement pattern of the first subject. In order to create a vision of that the second person is speaking, a picture of the speaker is applied to the model. The polygon model is at that modified to adapt to the second person. The to the model applied picture can consist of stills or moving pictures which have been stored in the database or have been transferred via, for instance, the telecommunication network. A three-dimensional picture is in this way created.
The registered movements in the first face consist of points which have been arranged in a three-dimensional face model. The face model consists of a number of polygons out together by points. The points consist of measuring points in the face of a subject, which points are registered during recording of sounds/polyphones. The registering of the points in the face of the subject is preferably made by marking of selected points in the face of the subject. The points after that are registered by means of, for instance, laser technology, and a bank over sounds and movement patterns is created.
The field of use of the invention is all cases where a reproduction of sound/speech shall be given a lifelike movement pattern in a speaking person's face. It can, for instance, relate to a person who is speaking a first language, but who by means of speech synthesis is represented speaking a second language. Such conditions should in the future be associated with telephony where the telecommunication system or equipments at the phoning persons translate the speech and represent the speaking persons in picture. The field of use of the invention, however, is not only telephony, but all connections where a first speech produced by a person shall be translated and reproduced in a second language with lifelike facial movements.
The indicated invention makes possible that a cheaper procedure can be applied at animation of speech with a belonging face. This is utilized, for instance, at translation of a speaking person's speech at translation from a first language to a second language. At recording is required only one subject who is utilized for production of the basic movements in the face. The person/persons who shall borrow the movement pattern from the subject need only record a number of sound sequences from which polyhones can be extracted. By registering polyphones and belonging facial movements of a suitable selection of persons, a bank can be created which can be utilized in different situations for animation of different faces. The registered faces can for instance relate to different persons of different ages and of different sex.
FIG. 1 shows how the facial movements are registered.
FIG. 2 shows how a polygon model is built up.
FIG. 3 shows a flow diagram of the method steps of the claimed invention.
In the following the invention is described on basis of the figures and the terms in them.
Registering of sounds with belonging movement patterns of the face of a subject is performed by utilization of, for instance, nonsense words, or from running text. Using nonsense words has advantages in that the subject at both production of words as well as facial expressions takes a neutral position. The nonsense words principally consist of one-syllable words which are divided into half-syllables. The half-syllables are linked together to words and sentences. Linking of two half-syllables is made in the middle of the vowel for respective half-syllable. The creating of a word is made according to the following; first the half-syllables which shall be part of the word, as well as the accentuation of the word in question, are found out. A single word, for instance “nod” (nu:d) is divided into the following syllables, “no”, “od”. Linking of “no” and “od” is after that made in the vowel “o”. From language dictionary, information is obtained about the lexical accent of the word. Said sounds after that are put together to words and sentences where the accent of each word is produced, as well as the sentence intonation of the whole sentence is determined. A build-up of words and sentences according to this procedure is previously known and described in i.a. patent document Se 9504367-5.
In order to create a polyphone bank, a subject is utilized who speaks a large number of words, for instance nonsense words. Alternatively also running text can be utilized. At the same time as said words are registered, also the facial movements of the subject are registered. These facial movements are stored in a movement base. The recorded words are divided into polyphones, half-syllables together with corresponding movements. A connection between the polyphones and corresponding movements is in this way created.
Registering of the facial movements is made by marking of different points in the face of the subject. The points are placed tighter where big changes are taking place, and with larger spaces in areas with smaller deviations in the movements. Areas with large deviations are for instance the area around the lips, and the movements of the lips. Each of the points is given a position in a three-dimensional reproduction of the subject. The picture only constitutes a polygon model of the face. By coordinating the model with polyphones which are put together to words and sentences, a movement pattern which corresponds to the produced one is obtained.
Polyphones are further registered from a second person, however without registering of the facial movements of the second person. Also in this case nonsense words are preferably used, but running text can also be utilized. A polyphone base is in this way created for the second person. In order to create a movement base for the second person, after that the polyphone bank for the first subject is utilized. Corresponding polyphones in the polyphone bank and the polyphone base are compared with regard to the duration of the sounds. Deviations in duration are registered for the polyphones, after which corresponding movements in the movement bank are modified. The modified movement is after that stored in a movement base. The movement base and the polyphone base are at that connected to each other.
When the second person shall be animated, words and phrases are produced by putting together half-syllables taken from the polyphones. At the same time a three-dimensional polygon model is built up with corresponding movement patterns which have been taken from the movement base. The created picture consists of a thread-model where the end points in the polygons represent the measuring points in the face of the first subject.
To create a vision of that the second person produces the produced speech, a picture of the second person is applied is to the polyphone model. The polygon model is modified in connection with that to adjust to facial forms, size of the mouth etc, of the second person. The produced reproduction is three-dimensional and is shown simultaneously with the produced speech, at which speech and facial movements well correspond. The indicated method, however, does not allow that the movement pattern quite corresponds to the second person's normal movement pattern, but is characterized by the movement pattern of the first person.
A person, B, is speaking a first language which shall be translated to a second language, or a text is given and shall be allotted the person B. For the person, B, a polyphone bank has been created, but no movement patterns have been registered. The for B registered polyphones are compared with corresponding polyphones in a polyphone base with associated movement patterns of a person, A. The exact length of the sound segments are found out both in the polyphone bank and the polyphone base. The deviations are registered and a model face is created where the movement patterns from A are utilized. The movements, however, are modified in relation to the deviation between the polyhones in the polyphone bank and the polyphone base. Sound segments which are longer in the polyphone bank than in the polyphone base result in that corresponding movement pattern in the face model is extended. For sound segments which are shorter in the polyphone bank than in the polyphone base, a corresponding reduction of corresponding movement pattern in the model face is made. A base with movement patterns associated to the polyphone bank is in this way created. At putting together polyphones from B's polyphone bank to words and sentences, a model face with movements taken from the base with movement patterns is produced. The produced movement patterns correspond to B's way of pronouncing corresponding parts of the polyphones, but with A's basic movement pattern. B's face after that can be applied to the model, at which B speaks with his/her own voice, but is allotted a movement pattern belonging to A, at which picture and sound correspond to each other and a lifelike experience is obtained.
The invention is not limited to the above described example of embodiment, or to following patent claims, but may be subject to modifications within the frame of the idea of invention.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4841575 *||Nov 14, 1986||Jun 20, 1989||British Telecommunications Public Limited Company||Image encoding and synthesis|
|US5657426 *||Jun 10, 1994||Aug 12, 1997||Digital Equipment Corporation||Method and apparatus for producing audio-visual synthetic speech|
|US5826234||Dec 5, 1996||Oct 20, 1998||Telia Ab||Device and method for dubbing an audio-visual presentation which generates synthesized speech and corresponding facial movements|
|US6208356 *||Mar 24, 1997||Mar 27, 2001||British Telecommunications Public Limited Company||Image synthesis|
|EP0624865A1||Apr 28, 1994||Nov 17, 1994||Telia Ab||Arrangement for increasing the comprehension of speech when translating speech from a first language to a second language|
|EP0664537A2||Oct 24, 1994||Jul 26, 1995||Telia Ab||Method and arrangement in automatic extraction of prosodic information|
|EP0710929A2||Nov 6, 1995||May 8, 1996||AT&T Corp.||Acoustic-assisted image processing|
|EP0778560A2||Oct 30, 1996||Jun 11, 1997||Telia Ab||Device and method at speech synthesis|
|GB2231246A||Title not available|
|GB2250405A||Title not available|
|WO1997034292A1||Feb 11, 1997||Sep 18, 1997||Telia Ab||Method and device at speech-to-speech translation|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7117223 *||Feb 14, 2002||Oct 3, 2006||Hitachi, Ltd.||Method of interpretation service for voice on the phone|
|US7209882 *||May 10, 2002||Apr 24, 2007||At&T Corp.||System and method for triphone-based unit selection for visual speech synthesis|
|US7369992||Feb 16, 2007||May 6, 2008||At&T Corp.||System and method for triphone-based unit selection for visual speech synthesis|
|US7933772||Mar 19, 2008||Apr 26, 2011||At&T Intellectual Property Ii, L.P.||System and method for triphone-based unit selection for visual speech synthesis|
|US7962345||Aug 23, 2008||Jun 14, 2011||International Business Machines Corporation||Speech-to-speech generation system and method|
|US8086456 *||Jul 20, 2010||Dec 27, 2011||At&T Intellectual Property Ii, L.P.||Methods and apparatus for rapid acoustic unit selection from a large speech corpus|
|US8315872||Nov 29, 2011||Nov 20, 2012||At&T Intellectual Property Ii, L.P.||Methods and apparatus for rapid acoustic unit selection from a large speech corpus|
|US8788268||Nov 19, 2012||Jul 22, 2014||At&T Intellectual Property Ii, L.P.||Speech synthesis from acoustic units with default values of concatenation cost|
|US9236044||Jul 18, 2014||Jan 12, 2016||At&T Intellectual Property Ii, L.P.||Recording concatenation costs of most common acoustic unit sequential pairs to a concatenation cost database for speech synthesis|
|US20030033312 *||Feb 14, 2002||Feb 13, 2003||Atsuko Koizumi||Method of interpretation service for voice on the phone|
|US20100286986 *||Jul 20, 2010||Nov 11, 2010||At&T Intellectual Property Ii, L.P. Via Transfer From At&T Corp.||Methods and Apparatus for Rapid Acoustic Unit Selection From a Large Speech Corpus|
|U.S. Classification||704/258, 375/E07.271, 704/E21.02, 704/276, 375/E07.084, 704/266|
|International Classification||H04N7/52, G06T13/40, G06T13/20, G10L21/10, G10L13/00, G06T9/00, H04N7/26|
|Cooperative Classification||G06T13/40, H04N19/20, G06T13/205, G10L2021/105|
|European Classification||H04N21/2368, H04N21/434A, H04N7/26J4T, G06T13/40, G06T13/20A|
|Mar 12, 2002||AS||Assignment|
|Sep 6, 2005||FPAY||Fee payment|
Year of fee payment: 4
|Sep 13, 2005||AS||Assignment|
Owner name: TELIASONERA AB, SWEDEN
Free format text: CHANGE OF NAME;ASSIGNOR:TELIA AB;REEL/FRAME:016769/0062
Effective date: 20021209
|Sep 28, 2005||AS||Assignment|
Owner name: DATA ADVISORS LLC, NEVADA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:TELIASONERA AB;REEL/FRAME:017089/0260
Effective date: 20050422
|Sep 26, 2006||AS||Assignment|
Owner name: DATA ADVISORS LLC, NEVADA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:TELIASONERA AB;TELIASONERA FINLAND OYJ;REEL/FRAME:018313/0371
Effective date: 20050422
|Sep 28, 2009||FPAY||Fee payment|
Year of fee payment: 8
|Feb 10, 2012||AS||Assignment|
Owner name: INTELLECTUAL VENTURES I LLC, DELAWARE
Free format text: MERGER;ASSIGNOR:DATA ADVISORS LLC;REEL/FRAME:027682/0187
Effective date: 20120206
|Dec 13, 2013||REMI||Maintenance fee reminder mailed|
|May 7, 2014||LAPS||Lapse for failure to pay maintenance fees|
|Jun 24, 2014||FP||Expired due to failure to pay maintenance fee|
Effective date: 20140507